In-Surely
Interactive insurance-policy QA system built with retrieval-augmented generation.
Why I Built This
Insurance policy documents are dense, repetitive, and difficult to navigate when someone needs a clear answer quickly. I built In-Surely to reduce that friction with retrieval-grounded Q and A over real policy text and tables. The project was motivated by a practical question: can we make policy understanding less intimidating without sacrificing precision.
What It Does
- Extracts text and tabular content from policy PDFs.
- Builds semantic embeddings for retrieval.
- Uses cache-aware query handling for faster repeated questions.
- Applies cross-encoder reranking to improve final context quality.
- Generates final responses with GPT-based synthesis.
Outcome
A full notebook-based RAG workflow that can be run in Colab and adapted to other document-heavy domains.
Links
- Code/notebook: In-Surely
- Colab entry point: Open in Colab